How WRL Was Built

WRL is a production web evidence service. It is also a documented experiment in LLM-driven software development.

The Setup

A human (Ben) defined the product, made business decisions, and steered direction. An LLM (Claude) wrote virtually all the code, tests, infrastructure configuration, and documentation. The orchestration used despicable-agents, a multi-agent framework where specialist agents (security, infrastructure, API design, frontend, etc.) debate approaches before implementation.

The Build

WRL went from first commit to production service in 10 days (March 13-22, 2026), across 60+ documented development phases:

• Act 1 (Solid Foundation): Core capture pipeline, authentication, per-tenant keys, CORS, rate limiting, security hardening.
• Act 2 (Evidence-Grade): RFC 3161 timestamps from DigiCert TSA, audit logging, production CD pipeline. Upgraded from "tamper-evident" to "independently verifiable."
• Act 3 (Infrastructure): MCP server, web UI, batch captures, scheduled captures, docs site, landing page, usage metering via Stripe, webhooks.

The Documentation

Every phase is documented in the evolution log: the prompts given, the agent debates, the decisions made, and the outcomes produced. When agents disagreed, both positions are recorded along with the resolution. When the human overrode an agent recommendation, the rationale is documented.

Browse the evolution log: docs/evolution/

Why Document This?

Most AI-assisted development happens behind closed doors. WRL's evolution log is an attempt to show what LLM-driven development actually looks like at production scale -- the parts that work, the parts that don't, and the human judgment calls that bridge the gap.

The product stands on its own merits regardless of how it was built. But if you're evaluating what LLMs can and can't do for software development, the evolution log is a primary source.